This paper addresses a personalized assistance control strategy for a powered hip exoskeleton, driven by a real-time human musculoskeletal (MSK) digital twin. The authors aim to develop an adaptive assistance controller that optimizes support by incorporating feedback on muscle properties from the MSK digital twin.
To enable real-time computation of the MSK, the authors used a muscle redundancy solver to calculate the model’s dynamics based on the user’s motion data. By mapping muscle space to joint space, they define the torque control of the hip exoskeleton robot based on the muscle moments estimated by the digital twin to deliver tailored assistance. Focusing on the energy cost of the hip joint flexor muscles at the onset of the swing phase of walking, the adaptive controller is built on a linear basis function of the moments generated by the rectus femoris and iliopsoas muscles, with dynamically adjusted coefficients.
The adaptive assistance coefficients are optimized in real-time using a gradient descent algorithm within a human-in-the-loop framework, minimizing the muscle forces. To validate the performance of the muscle redundancy solver, the authors compared the optimized muscle activations with those obtained by the computed muscle control (CMC) tools in OpenSim, confirming its accuracy and reliability. Experimental results from three healthy subjects show that the proposed controller significantly enhances walking performance. Compared to unassisted walking, it reduces estimated net normalized muscle forces by 10% (p < 0.001), increases hip flexion range of motion by 44% (p < 0.001) and decreases normalized hip joint torque by 14.7% (p < 0.001).
This paper introduces an innovative personalized adaptive assistance control system relying on a real-time MSK model, which is continuously updated through human-in-the-loop optimization.
